Free · Production AI Starter Kit

My AI agent follows 5 rules. Here's the file.

The CLAUDE.md template, the habit that turns mistakes into rules, and ten checks that catch AI systems failing quietly. One email unlocks all of it.

You'll also get the Thursday build log: one thing that broke, what it cost, and the fix. Unsubscribe from any email. Privacy.

The CLAUDE.md template, all 5 rules, ready to copy
The lab-notes loop and the retro prompt
10 canary checks, each from a real failure
A 15-minute setup and a first-week plan, plus the PDF

Start here · free preview

Three parts. Fifteen minutes.

Written for Claude Code. The same ideas work in Cursor, Codex or any agent that reads an instructions file such as AGENTS.md.

Part	What it does	Setup
CLAUDE.md template	Loads your rules into every session, and fires each rule only when its trigger is met	5 min
Lab-notes loop	Turns each real mistake into a 1–3 line rule the next session reads	2 min a lesson
Ten canary checks	Catch the failures that report success	One per system

Here's rule one, the only one that's always on:

CLAUDE.md · lines 7–11

## Always on: secrets hygiene
Never print, paste or echo the VALUE of a secret (API key, token, password).
Refer to secrets by variable name and file only, e.g. "OPENAI_API_KEY in .env".
Never commit .env files. Check .gitignore covers .env and .env.* and keeps .env.example.
If I paste a secret into chat, tell me it is now in the logs and must be rotated.

The full template, all 5 ruleslocked
The lab-notes loop and the retro promptlocked
Ten canary checkslocked
Your first week, day by daylocked
PDF and CLAUDE.md downloadslocked

Keep reading, free

Enter your email once to unlock the rest.

Unlock the kit

Rather have it built for you?

KET Digital builds and runs production AI systems for businesses: the agents, the monitoring, and the fixes when something breaks.

Book a call